In this paper, we discuss a new hybridmodel obtained by fusing a SARIMA model and a generalized singleneuronmodel. The proposed model has several advantages: ﬁ rst, it is able to the capture nonlinear behavior in the data; second, the SARIMA approach provides the modeler with a well-known and accepted methodology for model speciﬁ cation; and third, it is not necessary to use heuristics and expert knowledge for selecting the conﬁ guration of the artiﬁ cial neural network because the GSMN model uses the same inputs as the SARIMA model and it is not necessary to specify processing layers as in other neural network architectures. To assess the effectiveness of our model, we forecast the monthly demand of electricity in the Colombian energy market using several competitive models and we compare the accuracy of forecasts. The results obtained show that our approach performs better than the SARIMA and GSMN models in isolation. However, further research is needed to gain more conﬁ dence and to better understand the proposed model.
The hybrid genetic algorithm (modern) and local search (classical) is an example of hybrid method that can offer an opportunity to find a global optimum solution for electrical energy demandforecastingmodel. For most of the electrical energy demandforecasting models which use evolutionary algorithms (e.g. Genetic algorithms); the objective function can not obtain a good result because it can be trapped in a local optimum solution. This problem cannot be solved even though operating a single genetic algorithm repeatedly applied .
In the present study, a hybrid optimizing algorithm has been proposed using Genetic Algorithm (GA)and Particle Swarm Optimization (PSO) for Artificial Neural Network (ANN) to improve the estimation of electricitydemand of the state of Tamil Nadu in India. The GA- PSO model optimizes the coefficients of factors of gross state domestic product (GSDP) , electricity consumption per capita, income growth rate and consumer price index (CPI) that affect the electricitydemand. Based on historical data of 25 years from 1991 till 2015 , the simulation results of GA-PSO models are having greater accuracy and reliability than single optimization methods based on either PSO or GA. The forecasting results of ANN-GA-PSO are better than models based on single optimization such as ANN-BP, ANN-GA, ANN-PSO models. Further the paper also forecasts the electricitydemand of Tamil Nadu based on two scenarios. First scenario is the "as-it-is" scenario , the second scenario is based on milestones set for achieving goals of "Vision 2023" document for the state. The present research also explores the causality between the economic growth and electricitydemand in case of Tamil Nadu. The research indicates that a direct causality exists between GSDP and the electricitydemand of the state.
regulatory agencies, producers and consumers have a great interest in the study of electricity load and price. Since electricity cannot be stored, the demand must be satisfied instantaneously and producers need to anticipate to future demands to avoid overproduction. Good forecasting of electricitydemand is then very important for the agents in the market. In the past, demand was predicted in centralized markets  but competition has opened a new field of study. On the other hand, if producers and consumers have reliable predictions of electricity price, they can develop their bidding strategies and establish a pool bidding technique to achieve a maximum benefit. Consequently, prediction of electricitydemand and price are significant problems in this sector. In demand and price prediction two types of forecasts are considered: short term forecasts, this is one day ahead hourly forecasts, and mid term forecasts, i.e. several days ahead predictions for daily data.
A hybridmodel, which combines the seasonal time series ARIMA (SARIMA) and the multilayer feed- forward neural network to forecast time series with seasonality, is shown to outperform both two single models. Besides the selection of transfer functions, the determination of hidden nodes to use for the non linear model is believed to improve the accuracy of the hybridmodel. In this paper, we focus on the selection of the appropriate number of hidden nodes on the non linear model to forecast Malaysia load. Results show that by using only one hidden node, the hybridmodel of Malaysia load performs better than both single models with mean absolute percentage error (MAPE) of less than 1%. Keywords: Load Forecasting; Seasonal Autoregressive Integrated Moving Average; Multilayer Feed- forward Neural Network; HybridModel; Hidden Nodes.
The purpose of the project is to find or search the best model in order to modelling and forecasting student’s residential area electricitydemandusing statistical analysis. The best method that has been selected is using ARIMA. The ARIMA models and their versions have achieved a considerable success for electricitydemand. Then, by using the model for forecast the future demand for student’s residential area, ARIMA model can be use when the time series was stationary without missing the data. They can be further hybridized with artificial intelligence techniques. However, the complexity of demand pattern depends on its base period which it is changes from fairly smooth curve (annually based) to most noisy and cyclic complex curve (hourly based) since the effect of environmental factors increases. Combined forecasting was also introduced based on a few linear combination of various result from different forecasting methods.
We use an empirical fluctuation test based on the OLS-based CUSUM (cumulative sums). This is from the generalized fluctuation test framework, and involves fitting a constant model to the data and deriving an empirical process, which is intended to capture the fluctuations in the residuals. A null hypothesis of “no structural change” is assumed, and this is rejected when the fluctuations of the empirical process become improbably large compared to the fluctuations of the limiting process. Further details are provided in Zeileis et al. (2002). We test the null hypothesis that the model residuals and system load factors remain stationary over the years, by computing the OLS-based CUSUM process and plotting with standard and alternative boundaries.
The demand of electricity forms the basis for power system planning, power security and supply reliability. The need for forecasting models that evaluate the electric consumption with the highest level of accuracy is underlined by the black-outs for the whole Malaysia that occurred in 2005. The relevance of forecastingdemand for the utility company has become a much-discussed issue in the recent years which led to the development of new tools and methods for forecasting in the last two decades  . The issue of statistical forecasting versus non statistical forecasting or judgmental method of forecasting and decision making has been the focus of many debates for the past decades. It does become an issue too for Malaysian utility company in implementing their forecasting practices. The proponent of statistical techniques is stressing the importance of accuracy in forecast and consistency without the element of human variation and biasness. Bunn and Wright  explore the issues of quality of judgmental forecasts, judgmental adjustment of statistical forecasts and the practice of combining statistical approach and judgmental techniques for improving forecast accuracy. It is the current practice of utility company to employed short term forecast which is purely based on the expertise and experience of one forecaster. Through experience, the experts developed intuitive relationships between electrical load and weather parameters, time of day, day of week, season and time lag of response. Various factors need to be taken into account in order to arrive at hourly, daily and weekly forecast. These factors are daily temperature, legal and religious holidays, seasonal effects and human behavior whether they will take a day off preceding and following the holidays as to take advantage of a long break. Modifications in the electricity usage patterns are observed during these times as people have the tendencies of creating long weekend. Short term forecast based on the experienced forecaster is highly reliable with forecast error in the range of two to three percents. Lawrence and O’Connor [3,4] compared several statistical forecasting methods from naïve forecasts to an average judgment forecasts.
With the rapid growth of economy in China, domestic income and the number of domestic tourists increased rapidly. To accurately predict the number of tourists can keep the domestic tourism economy in good condition. Tourism demand and forecasting problems have been widely concerned by researchers in recent years. The Xi'an Museum has its unique cultural atmosphere and strong high-grade history, so attracts a lot of tourists. Therefore, more and more researchers pay more attention to the tourism demandforecasting problems. Construct a reasonable forecastingmodel for tourism demand, which helps human make accurate predictions about the tourism arrivals and make reasonable plans.
The Multi-Layer Perceptron (MLP) is one of the most widely used kinds of neural networks. It consists of a network based on multiple layers of perceptron-type neurons , trained by the back propagation technique (Sáenz & Ballesteros, 2002). The MLP generates a predictive model for one or more dependent variables (destination) based on the values of the predictors (SPSS, 2007). This is characterized by non-linearity in the output, layers of hidden neurons and a high degree of connectivity. It is supervised training and uses the algorithm of retro-error propagation. This algorithm is based on the rule of learning by error correction, considered as a generalization of the least squares algorithm (LMS), used in adaptive filtering through simple linear networks (Barbosa et al., 2001) and (Caparrini, 2015).
Overall, all these Turkish electricitydemand forecasts from the above studies foresaw electricitydemand being somewhat greater than it actually turned out to be. According to Keleş (2005), this is mainly due to “technical deficiencies of the models used, lack of ability of the relevant authorities in creating precise assumptions and not having transparency and accountability in the relevant processes” (p. vi). As a result of the these ‘exaggerated forecasts’, policies that were implemented (such as introducing the guaranteed BOO, BOT and TOOR projects) caused a significant proportion of electricity generation capacity of public power plants to remain idle, increasing the primary energy imports which Turkey does not need. These policies lead Turkey to be energy import dependent and therefore more vulnerable to external shocks and prevented energy markets from liberalizing (Keleş, 2005). Furthermore Ediger and Tatlidil (2002) stated that the values of the future predictions of demographic and economic variables used in the MAED models by SPO were significantly manipulated by government policies in line with high economic growth targets rather than reliable forecasts (Ediger and Tatlidil, 2002).
Meanwhile, ANN is a mathematical model used in many business applications for pattern recognition, forecasting, prediction and classification. This is due to its ability to “learn” from the data, its nonparametric nature and the ability to generalize . The second method used to forecast tourism demand in this study is by using artificial neural network. This method is a non-linear model that mimics human brain’s function, with the complex system of biological neurons model. ANNs demonstrate the capacity of improving time series forecasting through the analysis of additional information, reducing its size and lessening its complexity . ANN is also capable of applying learning process on sample data besides solving complicated and nonlinear forecast on tourist arrivals . As other methods faced difficulties, such as time-consuming and expensive, this intelligent ANN method handles the challenges well as it does not learn from past data. Other than that,  and ’s research on forecasting tourist arrivals in Hong Kong and South Africa had also concluded that neural network performs better than the other time series forecasting methods.
This study, which is the first of its kind in Zimbabwe, uses annual time series data on electricitydemand in Zimbabwe from 1971 to 2014, to model and forecast the demand for electricityusing the Box-Jenkins ARIMA framework. The study is guided by three objectives and these are: to analyze electricity consumption trends in Zimbabwe over the study period, to develop a reliable electricitydemandforecastingmodel for Zimbabwe based on the Box-Jenkins ARIMA technique and last but not least, to project electricitydemand in Zimbabwe over the next decade (2015 – 2025). Diagnostic tests indicate that X is an I (1) variable. Based on Theil’s U, the study presents the ARIMA (1, 1, 6) model, the diagnostic tests further show that this model is stable and hence suitable for forecastingelectricitydemand in Zimbabwe. The selected optimal model, the ARIMA (1, 1, 6) model proves beyond any reasonable doubt that in the next 10 years (2015 – 2025), demand for electricity in Zimbabwe will continue to fall. Amongst other policy recommendations, the study advocates for the liberalization of the electricity power sector in Zimbabwe in order to pave way for more efficient private investment whose potential is envisaged to adequately meet the existing demand for electricity.
A Naïve Benchmark was selected for this research to see how the forecasting methods compare. Based off of prior work in the area, a practical and easy to apply method was used. The Naïve Benchmark used (Equation 7) accounts for the seasonality by using averages. Although two seasonalities have been identified in the data set, this method only considers one seasonality. The within week seasonality was considered since it covers a longer period of time. The forecast is calculated by averaging the data for the corresponding half hours in the four previous weeks, and then adding in the error of the previous half hour.
This paper uses half-hourly electricitydemand data in South Australia as an empirical study of nonparametric modeling and forecasting methods for prediction from half-hour ahead to one year ahead. A notable feature of the univariate time series of electricitydemand is the presence of both intraweek and intraday seasonalities. An intraday seasonal cycle is apparent from the similarity of the demand from one day to the next, and an intraweek seasonal cycle is evident from comparing the demand on the corresponding day of adjacent weeks. There is a strong appeal in usingforecasting methods that are able to capture both seasonalities. In this paper, the forecasting methods slice a seasonal univariate time series into a time series of curves. The forecasting methods reduce the dimensionality by applying functional principal component analysis to the observed data, and then utilize an univariate time series forecasting method and functional principal component regression techniques. When data points in the most recent curve are sequentially observed, updating methods can improve the point and interval forecast accuracy. We also revisit a nonparametric approach to construct prediction intervals of updated forecasts, and evaluate the interval forecast accuracy.
The method of bootstrapping involves randomly resampling historical data. With time series, it is important to preserve any seasonal or trend patterns as well as the inherent serial cor- relation. The standard method for bootstrapping time series is the “block bootstrap” (Politis, 2003) which involves taking random segments of the historical time series and pasting them together to form new artificial series. There are obviously a very large number of such series that could be formulated in this way. A key parameter in the technique is the length of each segment. This needs to be long enough to capture the essential serial correlations in the data, but short enough to allow a large number of possible simulated series to be generated. When applied to seasonal time series, it is important that the length of each segment or block is a multiple of the length of the seasonal period. Politis (2001) calls this a “seasonal block bootstrap” although we will call it a “single season block bootstrap” to distinguish it from the double seasonal version to follow.
Specifically, in , the authors proposed methods including hybrid networks of self-organized map (SOM) and support-vector machine (SVM) to predict short-term electricity price. With the trained network, one can predict the future hourly elec- tricity prices in one day ahead. To confirm its feasibility, the proposed model had been trained and tested on the data of historical energy prices from the New Eng- land electricity market. In addition, in , a sensitivity analysis of similar days (SD) parameters to rise the accuracy of ANN model and SD-based short-term price fore- casting model were presented. In order to train the network, a large sum of data were used. The model had been tested in Pennsylvania-New Jersey-Maryland (PJM) elec- tricity market. The results showed that the mean absolute percentage error (MAPE) was around 11%. Furthermore, in , the authors introduced a method to predict next-day electricity prices based on the ARIMA methodology which was used to analyze the time series problem. The ARIMA model was tested in California elec- tricity market. More than 30-day historical data samples were required to train the model.
increasing energy efficiency . In recent years, these so- lution concepts started to pose new challenges to the existing power grids, whose hierarchical, centrally-controlled structure has remained unchanged for a century. For example, the exploitation of renewable sources such as solar or wind may be problematic due to their variable and intermittent nature , while the integration of distributed energy resources may cause congestion and atypical power flows that threaten system’s reliability . On the other hand, as a part of smart grid initiatives, smart meters have been widely deployed to under- stand the energy consumption behavior of the demand side. More specifically, it contains the information of how end users consume electricity in near real time. However, this also means that utility companies worldwide face challenges on managing big (smart meter) data on their hands of at least big volume, big velocity, and big value, whose benefits are waiting to be discovered .
The integrated energy management control algorithm is to take into account the specifications and functionality of the STATCOM, physical characteristics of the battery energy storage system, the extent at which peak demand is to be reduced, the instantaneous conditions of the LV distribution network and also electricitydemand forecasts. Given the continued trend towards smart-grid technology implementation, electricitydemand information in the future will be recorded throughout the network including high resolution residential and commercial electricitydemand data. The energy management control algorithm will be required to read electricitydemand data from each connected component, then in real-time formulate and continuously refine future electricitydemand forecasts in order to optimise energy resource allocation. This paper will cover recent progress on the development of an electricitydemand statistical modelling technique that will form the energy management control algorithm's forecasting component.
forecastingelectricity load demand is usually affected by other causal factors of data or disturbances, such as high frequency, non-stationary, non-constant variance and mean, and multiple seasonality, which are very likely related to half-hourly, hourly, daily, and weekly periodicity, and the calendar effects, for example, holidays and weekends. Therefore, modeling such data type poses multitude of challenges and the method must satisfy the causal factor that affects forecasting process. One of the methods that eliminate the causal factor of electricitydemand data is the EMD method. Then, it had been necessary to combine the EMD with the DR method, in order to improve forecast accuracy, rather than using a single method and also to investigate the elimination of causal factor in electricitydemand data.